Program

[Abstract]
　When estimating the covariance between two different asset returns with financial high frequency data, we need to simultaneously solve two different kind of problems: non-synchronous bias and market micro structure noise. Cumulative Covariance estimator was proposed by Hayashi and Yoshida (2005) to construct an unbiased estimator with non-synchronous data. Subsampling methods have been well known as a determined technique to reduce the variance of realized variance estimators under the situation where data are contaminated with the Market Microstructure Noise. The subsampling version of Cumulative Covariance estimator has been recognized as a possible candidate to handle these two problems through recent studies such as voev and Lunde (2007), Griffin and Oomen (2011). In this paper we examine the Susampling Cumulative Covariance (SCC) estimator as a special form of the Weighted Realized Covariance (WRC) proposed by Kanatani (2004). By minimizing the Mean Squared Error of the WRC we provide a framework to select the number of its subgrids which play a central role in the SCC estimator. Monte Carlo experiments show that the SCC estimator outperforms the other existing ones.